All labs

Missingness Mechanisms: MCAR, MAR, MNAR

Delete data three different ways and watch which repairs work, which fail, and why more data never fixes a biased mechanism.

Ch 2 2.3
Misconception
Foundations & Data Quality
18 mindifficulty 3/5

Controls

Missingness mechanism
Handling strategy
35%
600

Bias does not decrease with n — watch this while it slides.

Show the truth

Overlay the biomarker distribution before deletion.

Random seed

Every result on this page is a deterministic function of the seed and the controls.

Biomarker distribution

Analysed values versus the complete data that generated them.

00.050.100.150.2010203040506070[1.90, 4.11) n=0[4.11, 6.32) n=1[6.32, 8.53) n=0[8.53, 10.74) n=0[10.74, 12.95) n=1[12.95, 15.17) n=1[15.17, 17.38) n=4[17.38, 19.59) n=8[19.59, 21.80) n=9[21.80, 24.01) n=11[24.01, 26.22) n=21[26.22, 28.44) n=25[28.44, 30.65) n=32[30.65, 32.86) n=39[32.86, 35.07) n=262[35.07, 37.28) n=53[37.28, 39.49) n=55[39.49, 41.70) n=47[41.70, 43.92) n=31[43.92, 46.13) n=0[46.13, 48.34) n=0[48.34, 50.55) n=0[50.55, 52.76) n=0[52.76, 54.97) n=0[54.97, 57.18) n=0[57.18, 59.40) n=0[59.40, 61.61) n=0[61.61, 63.82) n=0[63.82, 66.03) n=0[66.03, 68.24) n=0[68.24, 70.45) n=0[70.45, 72.66) n=0true μestimatebiomarkerdensity
Analysed valuesComplete dataEstimated mean

Who went missing

Missing rows are highlighted at their imputed (or true) position.

102030405060700204060agebiomarker
ObservedImputed
True mean
39.734
Estimated mean
33.577
Bias
-6.157
systematically wrong
Analysed SD
5.316
true 10.522 — mean imputation shrinks it
Rows missing
210
Median analysed
33.577